Three-dimensional Homogeneous Isotropic Turbulence Research Articles

An Implicit Factorized Transformer with Applications to Fast Prediction of Three-dimensional Turbulence

Transformer has achieved remarkable results in various fields, including its application in modeling dynamic systems governed by partial differential equations. However, transformer still face challenges in achieving long-term stable predictions for three-dimensional turbulence. In this paper, we propose an implicit factorized transformer (IFactFormer) model, which enables stable training at greater depths through implicit iteration over factorized attention. IFactFormer is applied to large eddy simulation of three-dimensional homogeneous isotropic turbulence (HIT), and is shown to be more accurate than the FactFormer, Fourier neural operator (FNO), and dynamic Smagorinsky model (DSM) in the prediction of the velocity spectra, probability density functions of velocity increments and vorticity, temporal evolutions of velocity and vorticity root-mean-square value and isosurface of the normalized vorticity. IFactFormer can achieve long-term stable predictions of a series of turbulence statistics in HIT. Furthermore, IFactFormer showcases superior computational efficiency compared to the conventional DSM in large eddy simulation.

Departure from the statistical equilibrium of large scales in forced three-dimensional homogeneous isotropic turbulence

We study the statistically steady states of the forced dissipative three-dimensional homogeneous isotropic turbulence at scales larger than the forcing scale in real separation space. The probability density functions (p.d.f.s) of longitudinal velocity difference at large separations are close to, but deviate from, Gaussian, measured by their non-zero odd parts. The analytical expressions of the third-order longitudinal structure functions derived from the Kármán–Howarth–Monin equation prove that the odd-part p.d.f.s of velocity differences at large separations are small but non-zero. Specifically, when the forcing effect in the displacement space decays exponentially as the displacement tends to infinity, the odd-order longitudinal structure functions have a power-law decay with an exponent of $-$ 2, implying a significant coupling between large and small scales. Under the assumption that forcing controls the large-scale dynamics, we propose a conjugate regime to Kolmogorov's inertial range, independent of the forcing scale, to capture the odd parts of p.d.f.s. Thus, dynamics of large scales departs from the absolute equilibrium, and we can partially recover small-scale information without explicitly resolving small-scale dynamics. The departure from the statistical equilibrium is quantified and found to be viscosity-independent. Even though this departure is small, it is significant and should be considered when studying the large scales of the forced three-dimensional homogeneous isotropic turbulence.

Information production in homogeneous isotropic turbulence.

We study the Reynolds number scaling of the Kolmogorov-Sinai entropy and attractor dimension for three-dimensional homogeneous isotropic turbulence through the use of direct numerical simulation. To do so, we obtain Lyapunov spectra for a range of different Reynolds numbers by following the divergence of a large number of orthogonal fluid trajectories. We find that the attractor dimension grows with the Reynolds number as Re^{2.35} with this exponent being larger than predicted by either dimensional arguments or intermittency models. The distribution of Lyapunov exponents is found to be finite around λ≈0 contrary to a possible divergence suggested by Ruelle. The relevance of the Kolmogorov-Sinai entropy and Lyapunov spectra in comparing complex physical systems is discussed.

Size distribution spectrum of noninertial particles in turbulence.

Collision-coalescence growth of noninertial particles in three-dimensional homogeneous isotropic turbulence is studied. Smoluchowski's coagulation equation describes the evolution of the size distribution of particles in this system. By applying a methodology based on turbulence theory, the equation is shown to have a steady-state solution, which corresponds to the Kolmogorov-type power-law spectrum. Direct numerical simulations of turbulence and Lagrangian particles are conducted. The result shows that the size distribution in a statistically steady state agrees accurately with the theoretical prediction.

Wavelet-based regularization of the Galerkin truncated three-dimensional incompressible Euler flows.

We present numerical simulations of the three-dimensional Galerkin truncated incompressible Euler equations that we integrate in time while regularizing the solution by applying a wavelet-based denoising. For this, at each time step, the vorticity field is decomposed into wavelet coefficients, which are split into strong and weak coefficients, before reconstructing them in physical space to obtain the corresponding coherent and incoherent vorticities. Both components are multiscale and orthogonal to each other. Then, by using the Biot-Savart kernel, one obtains the coherent and incoherent velocities. Advancing the coherent flow in time, while filtering out the noiselike incoherent flow, models turbulent dissipation and corresponds to an adaptive regularization. To track the flow evolution in both space and scale, a safety zone is added in wavelet coefficient space to the coherent wavelet coefficients. It is shown that the coherent flow indeed exhibits an intermittent nonlinear dynamics and a k^{-5/3} energy spectrum, where k is the wave number, characteristic of three-dimensional homogeneous isotropic turbulence. Finally, we compare the dynamical and statistical properties of Euler flows subjected to four kinds of regularizations: dissipative (Navier-Stokes), hyperdissipative (iterated Laplacian), dispersive (Euler-Voigt), and wavelet-based regularizations.

Turbulent transport with intermittency: Expectation of a scalar concentration

Scalar transport by turbulent flows is best described in terms of Lagrangian parcel motions. Here we measure the Eulerian distance travel along Lagrangian trajectories in a simple point vortex flow to determine the probabilistic impulse response function for scalar transport in the absence of molecular diffusion. As expected, the mean squared Eulerian displacement scales ballistically at very short times and diffusively for very long times, with the displacement distribution at any given time approximating that of a random walk. However, significant deviations in the displacement distributions from Rayleigh are found. The probability of long distance transport is reduced over inertial range time scales due to spatial and temporal intermittency. This can be modeled as a series of trapping events with durations uniformly distributed below the Eulerian integral time scale. The probability of long distance transport is, on the other hand, enhanced beyond that of the random walk for both times shorter than the Lagrangian integral time and times longer than the Eulerian integral time. The very short-time enhancement reflects the underlying Lagrangian velocity distribution, while that at very long times results from the spatial and temporal variation of the flow at the largest scales. The probabilistic impulse response function, and with it the expectation value of the scalar concentration at any point in space and time, can be modeled using only the evolution of the lowest spatial wave number modes (the mean and the lowest harmonic) and an eddy based constrained random walk that captures the essential velocity phase relations associated with advection by vortex motions. Preliminary examination of Lagrangian tracers in three-dimensional homogeneous isotropic turbulence suggests that transport in that setting can be similarly modeled.

Impact of the Peterlin approximation on polymer dynamics in turbulent flows.

We study the impact of the Peterlin approximation on the statistics of the end-to-end separation of polymers in a turbulent flow. The finitely extensible nonlinear elastic (FENE) model and the FENE model with the Peterlin approximation (FENE-P) are numerically integrated along a large number of Lagrangian trajectories resulting from a direct numerical simulation of three-dimensional homogeneous isotropic turbulence. Although the FENE-P model yields results in qualitative agreement with those of the FENE model, quantitative differences emerge. The steady-state probability of large extensions is overestimated by the FENE-P model. The alignment of polymers with the eigenvectors of the rate-of-strain tensor and with the direction of vorticity is weaker when the Peterlin approximation is used. At large Weissenberg numbers, the correlation times of both the extension and of the orientation of polymers are underestimated by the FENE-P model.

Subgrid-scale modeling for implicit large eddy simulation of compressible flows and shock-turbulence interaction

We derive and analyze a model for implicit Large Eddy Simulation (LES) of compressible flows that is applicable to a broad range of Mach numbers and particularly efficient for LES of shock-turbulence interaction. Following a holistic modeling philosophy, physically sound turbulence modeling and numerical modeling of unresolved subgrid scales (SGS) are fully merged, in a manner quite different from that of traditional implicit LES approaches. The implicit subgrid model is designed in such a way that asymptotic consistency with incompressible turbulence theory is maintained in the low Mach number limit. Compressibility effects are properly accounted for by a novel numerical flux function, which can capture strong shock waves in supersonic flows and also ensures an accurate representation of smooth waves and turbulence without excessive numerical dissipation. Simulations of shock-tube problems, Noh's three-dimensional implosion problem, large-scale forced and decaying three-dimensional homogeneous isotropic turbulence, supersonic turbulent boundary layer flows, and a Mach = 2.88 compression-expansion ramp flow demonstrate the good performance of the SGS model; across this range of flows, predictions are in excellent agreement with theory, direct numerical simulations, and experimental reference data. Results for implicit LES of canonical shock-turbulence interaction are compared with results of explicit LES using the dynamic Smagorinsky model. The analysis shows that details of the numerical method used for shock capturing clearly outweigh the effect of different turbulence modeling strategies in explicit and implicit LES. The implicit LES model recovers the ideal 2nd-order grid convergence of shock-capturing errors that has been predicted using Rapid Distortion Theory. The dynamic Smagorinsky model in conjunction with a hybrid method that combines sixth-order central differences with a seventh-order weighted essentially non-oscillatory scheme yields turbulence statistics that are very similar to the implicit LES results. However, while the explicit LES requires a tailored high-order low-dissipative numerical method that applies numerical dissipation only in shock normal direction, no such ad hoc adjustments are necessary with the proposed implicit LES method.

One-dimensional energy spectra in three-dimensional incompressible homogeneous isotropic turbulence

The paper investigates the detailed features of one-dimensional energy spectra in three-dimensional isotropic turbulence, based on the exact solution of Karman-Howarth equation. Particular interest will be paid on the degree to which spectral scaling can lead the spectral data to be collapsed. The theory appears to be consistent with the wealth of experimental data (Comte-Bellot and Corrsin, 1971) at least in low Taylor microscale Reynolds number.

Geometrical Properties of Turbulent Dispersion

It is shown that the shape statistics of four correlated particles, known as a tetrad, in a direct numerical simulation of three-dimensional homogeneous isotropic turbulence agree very well with a simple diffusion equation with a separation-dependent eddy diffusivity. The latter is essentially an extension of Richardson's model for particle-pair dispersion to tetrads. It is also shown that the degree of elongation of the tetrad in the inertial subrange of a turbulentlike flow is controlled by the exponent, m, in an eddy diffusivity of the form K(r)[proportionality]r(m) where r is the interparticle separation, becoming more elongated as m increases in the range 0≤m<2.

Conditional vorticity budget of coherent and incoherent flow contributions in fully developed homogeneous isotropic turbulence

We investigate the conditional vorticity budget of fully developed three-dimensional homogeneous isotropic turbulence with respect to coherent and incoherent flow contributions. The coherent vorticity extraction based on orthogonal wavelets allows to decompose the vorticity field into coherent and incoherent contributions, of which the latter are noise-like. The impact of the vortex structures observed in fully developed turbulence on statistical balance equations is quantified considering the conditional vorticity budget. The connection between the basic structures present in the flow and their statistical implications is thereby assessed. The results are compared to those obtained for large- and small-scale contributions using a Fourier decomposition, which reveals pronounced differences.

Wavelet Analysis of the Conditional Vorticity Budget in Fully Developed Homogeneous Isotropic Turbulence

We study the conditional balance of vortex stretching and vorticity diffusion of fully developed three-dimensional homogeneous isotropic turbulence with respect to coherent and incoherent flow contributions. This decomposition is achieved by the Coherent Vorticity Extraction based on orthogonal wavelets applied to DNS data, which yields insights into the influence of the different contributions as well as their interaction.

A Memory-Saving Algorithm for Spectral Method of Three-Dimensional Homogeneous Isotropic Turbulence

AbstractHomogeneous isotropic turbulence has been playing a key role in the research of turbulence theory. And the pseudo-spectral method is the most popular numerical method to simulate this type of flow fields in a periodic box, where fast Fourier transform (FFT) is mostly effective. However, the bottle-neck in this method is the memory of computer, which motivates us to construct a memory-saving algorithm for spectral method in present paper. Inevitably, more times of FFT are needed as compensation. In the most memory-saving situation, only 6 three-dimension arrays are employed in the code. The cost of computation is increased by a factor of 4, and that 38 FFTs are needed per time step instead of the previous 9 FFTs. A simulation of isotropic turbulence on 20483 grid can be implemented on a 256G distributed memory clusters through this method.

A mixed large eddy simulation model based on the residual-based variational multiscale formulation

A mixed model based on the residual-based variational multiscale (RBVM) formulation is proposed for the large eddy simulation of turbulent flows. In this model the cross stresses are modeled by the RBVM term and the Reynolds stresses are represented by a Smagorinsky eddy viscosity. The mixed model is motivated by an a priori analysis of a turbulent flow field that indicates that the RBVM term correctly models the dissipation induced by the cross-stress term but underpredicts the contribution from the Reynolds stress term. The dynamic version of the mixed model is implemented in a Fourier-spectral code in order to predict the decay of incompressible, three-dimensional homogeneous isotropic turbulence. It is found that the mixed model yields better agreement with direct numerical simulation solution than the dynamic Smagorinsky and the RBVM models, which are its building blocks.

Wavelet-based statistics of energy transfer in high-resolution direct numerical simulation of three-dimensional homogeneous isotropic turbulence

Small-scale statistics of wavelet-based quantities of turbulence are examined, with particular attention to t α,β[ ι ], which is the wavelet-based transfer of the wavelet-based energy e α[ ι ] at location ι and scale r α to all scales smaller than r β (⩽r α). A divergence-free orthonormal wavelet decomposition method is applied to direct numerical simulation data of three-dimensional incompressible homogeneous isotropic turbulence with grid points and Taylor microscale Reynolds numbers up to 20483 and 732, respectively. We propose a new statistical measure for the degree of scale locality of wavelet-based energy flux to scales smaller than r β constructed from t α,β[ ι ]. At R λ = 732, the measure retains a constant value in the inertial subrange, a value about two-thirds as large as that in the dissipation range, though the flux is local in scale throughout the range, from the inertial subrange to the dissipation range. We confirm that the influence of energy transfer from the large-scale flow on the flux becomes smaller with decreasing scales and negligible in the dissipation range. Scale-dependent spatial correlation between t α,β[ ι ] and e α[ ι ] is discussed. We find that, when r β is in the inertial range, the conditional correlation between forward transfer, which is t α,β[ ι ] at the positions satisfying t α,β[ ι ]>0, and e α[ ι ] becomes stronger as scale r α increases from r β, while this behavior is reversed when r β is in the dissipation range.

Coherent vortices in high resolution direct numerical simulation of homogeneous isotropic turbulence: A wavelet viewpoint

Coherent vortices are extracted from data obtained by direct numerical simulation (DNS) of three-dimensional homogeneous isotropic turbulence performed for different Taylor microscale Reynolds numbers, ranging from Reλ=167 to 732, in order to study their role with respect to the flow intermittency. The wavelet-based extraction method assumes that coherent vortices are what remains after denoising, without requiring any template of their shape. Hypotheses are only made on the noise that, as the simplest guess, is considered to be additive, Gaussian, and white. The vorticity vector field is projected onto an orthogonal wavelet basis, and the coefficients whose moduli are larger than a given threshold are reconstructed in physical space, the threshold value depending on the enstrophy and the resolution of the field, which are both known a priori. The DNS dataset, computed with a dealiased pseudospectral method at resolutions N=2563,5123,10243, and 20483, is analyzed. It shows that, as the Reynolds number increases, the percentage of wavelet coefficients representing the coherent vortices decreases; i.e., flow intermittency increases. Although the number of degrees of freedom necessary to track the coherent vortices remains small (e.g., 2.6% of N=20483 for Reλ=732), it preserves the nonlinear dynamics of the flow. It is thus conjectured that using the wavelet representation the number of degrees of freedom to compute fully developed turbulent flows could be reduced in comparison to the standard estimation based on Kolmogorov’s theory.

A Study of Turbulence Simulations Using Finite Difference Lattice Boltzmann Method

Three-dimensional homogeneous isotropic turbulence is simulated using the finite difference lattice Boltzmann method (FDLBM). First, we analyze the characteristics of numerical viscosity of FDLBM, and present a method in which the numerical viscosity coincides with that of the Navier-Stokes based FDM. Second, we conducted simulations with the Smagorinsky model. In the energy spectrum E(k) one detects the inertial range E(k) ∼k-5/3; however, the large k range presents a persistent energy “tail,” which has also been found in direct simulations using conventional lattice Boltzmann scheme. This energy “tail” is is due to generation of non-physical longitudinal noise. However, for high Reynolds number flows this noise is shown to be negligible.

The Study on Simulations of Turbulence by the Finite Difference Lattice Boltzmann Method

Three-dimensional homogeneous isotropic turbulence are simulated by the Finite Difference Lattice Boltzmann Method (FDLBM). Firstly, we analyzed the characteristics of numerical viscosity of FDLBM, and we present a method that the numerical viscosity coincides with that of the Navier-Stokes based FDM. Secondly, we conducted simulations with the Smagorinsky Model. In theenergy spectrum E (k) one detects the inertial range E (k) -k-5/3 but the large k range presents a persistent energy tail, which has been found in the case of direct simulations by conventional lattice Boltzmann scheme too. This energy tail is shown to be due to generation of non-physical longitudinal noise. But for high Reynolds number flows this noise is shown to be negligibly small.

Multiparticle dispersion in fully developed turbulence

The statistical geometry of dispersing Lagrangian clusters of four particles (tetrahedra) is studied by means of high-resolution direct numerical simulations of three-dimensional homogeneous isotropic turbulence. We give evidence of a self-similar regime of shape dynamics characterized by almost two-dimensional, strongly elongated geometries. The analysis of four-point velocity-difference statistics and orientation shows that inertial-range eddies typically generate a straining field with a strong extensional component aligned with the elongation direction and weak extensional/compressional components in the orthogonal plane.

An adaptive local deconvolution method for implicit LES

The adaptive local deconvolution method (ALDM) is proposed as a new nonlinear discretization scheme designed for implicit large-eddy simulation (ILES) of turbulent flows. In ILES the truncation error of the discretization of the convective terms functions as a subgrid-scale model. Therefore, the model is implicitly contained within the discretization, and an explicit computation of model terms becomes unnecessary. The discretization is based on a solution-adaptive deconvolution operator which allows to control the truncation error. Deconvolution parameters are determined by an analysis of the spectral numerical viscosity. An automatic optimization based on an evolutionary algorithm is employed to obtain a set of parameters which results in an optimum spectral match for the numerical viscosity with theoretical predictions for isotropic turbulence. Simulations of large-scale forced and decaying three-dimensional homogeneous isotropic turbulence show an excellent agreement with theory and experimental data and demonstrate the good performance of the implicit model. As an example for transitional flows, instability and breakdown of the three-dimensional Taylor–Green vortex are considered. The implicit model correctly predicts instability growth and transition to developed turbulence. It is shown that the implicit model performs at least as well as established explicit models.